Comprehensive evaluation of precipitation datasets over Iran

Abstract

Global gridded precipitation datasets have been developed using rain gauges, satellite observations, and data assimilation techniques to fulfill the need in regions with a limited contribution of ground observations like Iran. This study presents a comprehensive evaluation of currently available precipitation datasets over Iran at monthly (44 datasets) and daily (34 datasets) time scales. To include the maximum number of datasets and in situ data, we consider two periods for the evaluations, namely 2003–2010 for the daily and monthly assessment and 2014–2018 for the daily. For the assessment, a network of more than 1500 rain gauges is utilized within 2003–2010 and 370 rain gauges within 2014–2018. Moreover, we compare the pixel-to-pixel (interpolated in situ data v.s. gridded datasets), and point-to-pixel (in situ data as a point v.s. gridded datasets) approaches in assessing datasets performances. In terms of the Kling-Gupta efficiency (KGE) parameters, the datasets perform worse in bias at monthly time scales and correlation at daily time scales. However, considering in situ precipitation above 5 mm/day, all datasets perform poorly in estimating precipitation variability. We find that, in general, reanalysis products have a higher KGE, ranging between 0.41 (0.21) and 0.91 (0.71), than satellite-based products with a KGE ranging from 0.14 (-0.57) to 0.92 (0.57) over Iran at monthly (daily) scale. Moreover, GPCC overall matches the validation dataset better than other products over Iran’s basins, whereas CPC, ERA5, and IMERG-Final are more suitable for near-real-time studies. Also, if latency is a top criterion, PERSIANN-PDIR will be the first option. Indeed, PERSIANN-PDIR with a KGE value of 0.69 (0.33) at monthly (daily) time scale within 2003–2010 performs remarkably well, as a non-adjusted real-time satellite-based product. The comparison between the point-to-pixel and the pixel-to-pixel approaches shows that the point-to-pixel approach underestimates the quality of the datasets but does not change the ranking of the datasets.